Policy Implications and Conclusions
Appendix 4.2. Estimation Approach and Robustness Checks
global financial conditions, such as the J.P. Morgan Emerging Markets Bond Index (EMBI) spread and yield. There is much more cross-economy hetero-geneity in the correlation between domestic growth and the U.S. federal funds rate and the 10-year U.S.
Treasury bond rate. On average, only half of the sample shows a negative correlation between domes-tic growth and U.S. interest rates.
Appendix 4.2. Estimation Approach and
log diferences of the relevant level’s time series. he irst ive variables constitute the “external” or for-eign block, and the remaining variables make up the
“internal” or domestic block.
Identiication (the mapping to the structural shocks) uses contemporaneous restrictions on the structure of the matrix A0. he key restriction is that shocks to the external block are assumed to be exogenous to shocks to the internal block; in other words, the external variables do not respond to the internal variables con-temporaneously. Within the external block, structural shocks are further identiied using a recursive (Cho-lesky) scheme, deined by the ordering of the variables in the vector yt. herefore, U.S. real GDP growth is assumed to respond to other shocks only with a lag. U.S. inlation is afected by U.S. growth shocks contemporaneously, but by other shocks with a lag.
he U.S. interest rate responds contemporaneously to U.S. real GDP growth and inlation shocks, but not to the EMBI Global yield or to any emerging market economy’s terms-of-trade growth. he EMBI Global yield is placed ahead of economy-speciic terms-of-trade growth, but behind all the U.S. variables. Finally, terms-of-trade growth is placed last in the recursive ordering, implying that it responds contemporaneously to all other external variables, but not to the domestic variables. Structural shocks within the internal block are unidentiied.
All variables enter the model with four lags. Other than the contemporaneous restrictions on the matrix A0,
there are no restrictions on the coeicients for the lagged variables; that is, the lags of the internal block variables are allowed to afect the external block variables.
Estimation by Bayesian Methods
he number of sample observations relative to the number of parameters to be estimated in each equation of each economy’s SVAR is not very large. his means that there is a danger of overitting if the model esti-mation is left unrestricted. Overitting leads to good performance of the estimated model within the sample (as it tends to follow the noise in the sample more closely), but to poor out-of-sample performance.
here are a number of ways to address this overit-ting problem. One is to impose hard restrictions on the parameters, by ixing some of them to speciic values. However, by taking a hard stance before the fact, such restrictions rule out potentially interest-ing dynamics. An alternative to such restrictions is to estimate the model using Bayesian methods, which is the approach followed in this chapter. his involves specifying restrictions on estimated parameters that are softer, such as constraining them to be more likely at some values than at others. Operationally, a prior probability distribution is imposed on the estimated parameters, pulling in additional information from outside the sample observations, to avoid overitting.
his is combined with the information in the sample to generate estimates for the parameters.
Table 4.6. Correlations of Domestic Real GDP Growth with Key Variables, 1998–2013
U.S. Real GDP Growth
U.S. Federal Funds Rate
Ten-Year U.S. Treasury
Bond Rate
Euro Area Real GDP Growth
China Real
GDP Growth EMBI Spread EMBI Yield
Terms-of-Trade Growth
Argentina 0.12 –0.13 –0.28 0.15 0.56 –0.68 –0.64 0.33
Brazil 0.15 0.03 0.03 0.42 0.51 –0.51 –0.37 0.63
Chile 0.31 –0.01 –0.11 0.44 0.25 –0.62 –0.52 0.33
China –0.10 0.05 –0.05 0.16 1.00 –0.64 –0.50 –0.27
Colombia –0.08 –0.18 –0.28 0.15 0.53 –0.82 –0.71 0.29
India 0.27 0.10 0.19 0.42 0.66 –0.44 –0.29 0.03
Indonesia –0.32 –0.38 –0.35 –0.15 0.27 –0.56 –0.52 –0.26
Malaysia 0.26 –0.07 0.00 0.33 0.21 –0.37 –0.26 0.29
Mexico 0.76 0.35 0.18 0.77 0.16 –0.26 –0.16 0.52
Philippines 0.18 –0.27 –0.32 0.16 0.32 –0.61 –0.58 –0.40
Poland 0.40 0.44 0.36 0.61 0.49 –0.32 –0.13 –0.20
Russia 0.45 0.30 0.31 0.66 0.21 –0.23 –0.04 0.77
South Africa 0.39 0.32 0.23 0.67 0.42 –0.38 –0.18 –0.14
Thailand 0.17 –0.15 –0.07 0.18 0.26 –0.31 –0.24 0.15
Turkey 0.44 –0.06 –0.04 0.45 0.38 –0.51 –0.41 –0.14
Venezuela 0.17 0.12 –0.02 0.24 0.26 –0.48 –0.38 0.09
Source: IMF staff calculations.
Note: Period is 1998:Q1–2013:Q2. EMBI = J.P. Morgan Emerging Markets Bond Index.
each variable is assumed to follow a irst-order autore-gressive (AR(1)) process with independent, normally distributed errors. Given that the variables have already been transformed to induce stationarity, a random walk, with a unit AR(1) coeicient for the prior, would not be appropriate. Simple AR(1) regressions, however, do suggest estimated AR(1) coeicients of about 0.8, which is the AR(1) coeicient used in the prior for the baseline estimation. Some of this persistence relects the fact that all growth rates are calculated as year-over-year diferences.
he weight of the prior versus the sample in the estimation is determined according to the Bayesian approach presented in Sims and Zha (1998). If twice the number of parameters to be estimated in an equa-tion is greater than the estimaequa-tion sample size, the chapter applies a rule of thumb that gives the prior a (T – p)
relative weight of
1 – ————
∈ [0,1], in which 2(kp + 1)T is the number of available sample observations and k and p are deined as above.25
Figure 4.16 compares the average baseline SVAR results using the AR(1) priors with those from an alternative white-noise prior. As expected, with a white-noise prior, the impulse responses show lower persistence and amplitude. he conditional out-of-sample forecasts from these speciications are largely similar to those shown in Figures 4.12 and 4.13, although the forecast performance improves with a less persistent prior for some economies (for example, Malaysia, Mexico, and the Philippines).
Robustness of the Baseline Results
A variety of alternative speciications are used to assess the robustness of the main results. In particular, a number of additional variables are introduced as prox-ies for external demand, U.S. monetary policy, external inancing conditions, and the terms of trade. he results are described in the following.
25In the case of China, there are 60 observations for the reduced-form VAR. With 37 coeicients to estimate, the priors receive a weight (importance) of slightly less than 0.25 in the baseline speciication (and a maximum weight of 0.50 in the speciication for out-of-sample forecasting reported in the chapter text).
Alternative U.S. monetary policy measures
As described in the chapter, alternative proxies for global inancing conditions are tried to assess the robustness of the indings: the 10-year U.S. Treasury bond rate, which is in the baseline speciication (see Figure 4.16);
and alternative speciications in which the 10-year U.S.
Treasury bond rate is replaced by (1) the U.S. efec-tive federal funds rate; (2) the ex ante U.S. real federal funds rate; (3) the change in the U.S. federal funds rate;
(4) the U.S. term spread (deined as the 10-year U.S.
Treasury bond rate minus the U.S. federal funds rate);
(5) Kuttner (2001)–style unanticipated monetary policy shocks, inferred from the behavior of federal funds futures; and (6) an extension of the Romer and Romer (2004) exogenous monetary policy shock series, based on Coibion (2012).
–0.01 0.00 0.01 0.02 0.03 0.04 0.05
0 5 10 15 20
–0.4 –0.3 –0.2 –0.1 0.0 0.1 0.2
0 5 10 15 20
–0.2 –0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6
0 5 10 15 20
(Percentage points)
Baseline specification: AR(1) prior, ρ = 0.8 Alternative specification: white-noise prior
4. Terms-of-Trade Growth Shock 1. U.S. GDP Growth
Shock
2. U.S. Treasury Bond Rate Shock
3. EMBI Spread Shock
–0.6 –0.4 –0.2 0.0 0.2 0.4 0.6 0.8
0 5 10 15 20
Source : IMF staff calculations.
Note: AR(1) = first-order autoregression; EMBI = J.P. Morgan Emerging Markets Bond Index. Shocks are normalized to a 1 percentage point increase. X-axis units are quarters; t = 0 denotes the quarter of the shock.
Note that an increase in the U.S. federal funds or policy rate—nominal or real—negatively afects emerging market economies’ growth only after a lag of six quarters just as the 10-year U.S. Treasury bond rate does (Figures 4.17 and 4.18). he impact efect is negative for very few economies (Chile, Malaysia, hailand, Venezuela). hese puzzling results may indi-cate that the U.S. rate increase embodies expectations of an improvement in future U.S. growth. Indeed, even U.S. growth is adversely afected with a delay (see Table 4.1). Emerging market economies’ growth declines only as domestic interest rates gradually rise in response to the U.S. rate increase.
he alternative proxy using the term spread pro-duces a more immediate negative efect (Figure 4.17).
It is possible that the Federal Reserve’s heavy reliance on unconventional policies to lower long-term rates
over the past few years means that long-term rates are now a better measure of its stance than short-term rates. With the short-short-term rate at the zero lower bound, positive shocks to the term spread would indi-cate a tighter U.S. monetary policy (see also Ahmed and Zlate, 2013). With the exception of the U.S.
term spread, emerging markets’ growth responses to shocks to the alternative measures are similar to their responses to shocks to the 10-year U.S. Treasury bond rate or the U.S. policy rate.26
It is important to note that shocks to the 10-year U.S. Treasury bond rate may not correspond closely to unanticipated U.S. monetary policy changes unrelated to U.S. GDP growth and inlation. Because it is a long-term rate, it is much more likely that shocks to the 10-year rate relect expectations in regard to the U.S. economy. Furthermore, since the global inancial crisis, the 10-year U.S. Treasury bond rate has been suppressed by safe haven lows into U.S. Treasuries, relecting not just the U.S. growth outlook, but also uncertainty over the global recovery. herefore, shocks to the 10-year U.S. Treasury bond rate could occur in response to a wide range of external (non-U.S.) factors.
he impulse responses from speciications (5) and (6) use monetary policy measures to represent more accurately true U.S. monetary policy shocks. As shown in Figure 4.19, the sign and shape of the responses are broadly the same as for the other proxies discussed ear-lier. Growth in emerging market economies responds to U.S. monetary policy shocks only after one year.
he reason for such responses could be that monetary policy shocks have been fairly limited and muted over the sample period. As Figure 4.20 shows, the largest shocks are shown to have occurred in the 1980s, when calculated using the technique set out in Romer and Romer (2004), and to have occurred with much less frequency, when calculated using the information con-tained in federal funds futures contracts, as described in Kuttner (2001).
External inancing conditions
Robustness checks are also conducted for diferent types of external inancing shocks besides the EMBI Global yield used in the baseline speciication. he
26Another alternative speciication is also tried in which the 10-year U.S. Treasury bond rate is added after the policy rate in the external block. Shocks to either the policy rate or the 10-year rate in this expanded speciication still elicit a lagged negative growth response for most emerging markets.
–3 –2 –1 0 1 2 3
0 5 10 15 20
–0.6 –0.4 –0.2 0.0 0.2 0.4 0.6 0.8
0 5 10 15 20
–0.8 –0.6 –0.4 –0.2 0.0 0.2 0.4 0.6 0.8 1.0
0 5 10 15 20
Figure 4.17. Average Impulse Responses to Shocks from Alternative U.S. Monetary Policy Variables
(Percentage points)
U.S. federal funds rate U.S. real short-term rate U.S. term spread Change in U.S. federal funds rate
4. Domestic Real Exchange Rate 1. Domestic GDP Growth 2. U.S. GDP Growth
3. Domestic Short-Term Interest Rate
–0.8 –0.6 –0.4 –0.2 0.0 0.2 0.4 0.6 0.8
0 5 10 15 20
Source: IMF staff calculations.
Note: Shocks are normalized to a 1 percentage point increase. X-axis units are quarters; t = 0 denotes the quarter of the shock.
–0.8 –0.6 –0.4 –0.2 0.0 0.2 0.4
0 5 10 15 20
–0.3 –0.2 –0.1 0.0 0.1 0.2 0.3
0 5 10 15 20
–0.4 –0.2 0.0 0.2 0.4 0.6 0.8
0 5 10 15 20
–1.5 –1.0 –0.5 0.0 0.5 1.0 1.5 2.0
0 5 10 15 20
–0.8 –0.4 0.0 0.4 0.8
0 5 10 15 20 –0.8
–0.4 0.0 0.4 0.8 1.2
0 5 10 15 20 –0.6
–0.3 0.0 0.3 0.6 0.9 1.2
0 5 10 15 20
–0.3 0.0 0.3 0.6
0 5 10 15 20 –1.5
–1.0 –0.5 0.0 0.5 1.0 1.5
0 5 10 15 20 –3
–2 –1 0 1 2 3
0 5 10 15 20
–0.8 –0.4 0.0 0.4 0.8
0 5 10 15 20
–0.4 –0.2 0.0 0.2 0.4 0.6 0.8
0 5 10 15 20
–0.6 –0.3 0.0 0.3 0.6 0.9 1.2
0 5 10 15 20
–0.6 –0.3 0.0 0.3 0.6 0.9
0 5 10 15 20
–0.6 –0.3 0.0 0.3 0.6 0.9 1.2
0 5 10 15 20
–1.5 –1.0 –0.5 0.0 0.5 1.0 1.5 2.0 2.5
0 5 10 15 20
Source: IMF staff calculations.
Note: Shocks are normalized to a 1 percentage point increase. X-axis units are quarters; t = 0 denotes the quarter of the shock.
U.S. federal funds rate Ten-year U.S. Treasury bond rate
6. India
9. Mexico 10. Philippines
13. South Africa 14. Thailand
1. Argentina 2. Brazil
7. Indonesia 8. Malaysia
11. Poland 12. Russia
15. Turkey 16. Venezuela
3. Chile 4. China
5. Colombia (Percentage points)
variables used across the alternative speciications are (1) the EMBI Global spread and (2) the U.S.
high-yield spread. As Figure 4.21 shows, the average response of domestic GDP growth in the 16 emerging market economies to all three identiied shocks is very similar.
External demand conditions
he analysis assesses whether and how the efects of U.S. real GDP growth on domestic growth are afected by controlling for real GDP growth in the euro area.
he euro area growth indicator enters the external block of the SVAR after U.S. real GDP growth in the recursive identiication, but before the other U.S. vari-ables. However, placing euro area growth after all the
As shown in panel 1 of Figure 4.22, the average response of domestic growth to U.S. real GDP growth is largely unafected by the introduction of this addi-tional variable. Moreover, the response of domestic real GDP growth to euro area growth is also as strong as the response to U.S. real GDP growth, conirming that it is reasonable to use U.S. real GDP growth as a proxy for general advanced economy real growth shocks (Figure 4.22, panel 2). Some economy-speciic difer-ences appear in the results: for instance, economies with deeper external trade ties with the euro area (for example, Poland and South Africa) show larger growth efects with respect to euro area real GDP growth changes than with respect to U.S. real GDP growth changes, whereas growth in Mexico shows the reverse (that is, larger efects with respect to U.S. real GDP growth changes).
he analysis also considers China’s real investment growth as an alternative proxy (instead of China’s real GDP growth) for external demand shocks emanat-ing from China (Figure 4.22, panel 3). Although the pattern of domestic growth responses to changes in China’s investment growth is very similar to responses
–1.5 –1.0 –0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0
–2 –1 0 1 2 3 4
0 5 10 15 20
–0.6 –0.3 0.0 0.3 0.6 0.9 1.2 1.5 1.8
–2 –1 0 1 2 3 4 5 6
0 5 10 15 20
–0.6 –0.3 0.0 0.3 0.6 0.9 1.2 1.5
–2 –1 0 1 2 3 4 5
0 5 10 15 20
1. Domestic Real GDP Growth
Figure 4.19. Average Impulse Responses to Alternative Measures of U.S. Monetary Policy Shock
(Percentage points)
2. U.S. Real GDP Growth Based on Romer and Romer (2004)1 (left scale) Based on Kuttner (2001) (right scale)
–4 –2 0 2 4
–15 –10 –5
0 5 10 15
0 5 10 15 20
3. Domestic Short-Term Interest Rate
4. Domestic Real Exchange Rate
Sources: Federal Reserve Economic Data; Haver Analytics; IMF, International Financial Statistics database; Thomson Reuters Datastream; and IMF staff calculations.
Note: Shocks are normalized to a 1 percentage point increase. X-axis units in panels are quarters; t = 0 denotes the quarter of the shock.
1 See Coibion (2012).
–5 –4 –3 –2 –1 0 1 2 3 4 5
–0.4 –0.3 –0.2 –0.1 0.0 0.1 0.2 0.3 0.4
1969:
Q1
75 80 85 90 95 2000 05 08 13:
Q4 Source: IMF staff calculations.
Note: X-axis units in panels are quarters; t = 0 denotes the quarter of the shock.
1See Coibion (2012).
Figure 4.20. Alternative Monetary Policy Shocks (Percentage points)
Approach based on Romer and Romer (2004)1(left scale) Approach based on Kuttner (2001) (right scale)
to China’s real GDP growth, the elasticity is negligible on impact, building up slightly over time.
Terms-of-trade growth alternatives
As a potentially more exogenous proxy for emerging market economies’ terms of trade, the exercise includes the global commodity price index in the external block, placing it in the second position within the recursive ordering for the identiication of external structural shocks. Panel 4 of Figure 4.22 shows a simi-lar pattern of response to that resulting from a positive shock to terms-of-trade growth.
Longer time period
he economy-speciic SVARs are also estimated using the longest available quarterly data. Only three econo-mies have all baseline variables available from the irst quarter of 1995: Brazil, Mexico, and South Africa.
he results for those economies with additional data are not afected by the longer-sample SVAR. Figure
responses of domestic GDP growth to shocks from four of the key external factors. Similar results are obtained for Mexico and South Africa.
Robustness checks with panel vector autoregressions he inal section of this appendix assesses how the estimated relationship between emerging market economies’ growth and external conditions is afected by an alternative estimation technique in a panel setup.
A panel VAR allows for many more degrees of freedom relative to the SVAR because all the economy-speciic observations are pooled. As such, it provides a sense of the average behavior among the sample of economies
–0.6 –0.5 –0.4 –0.3 –0.2 –0.1 0.0 0.1 0.2
0 2 4 6 8 10 12 14 16 18 20
(Percentage points)
Response to EMBI yield Response to EMBI spread Response to U.S. high-yield spread
Sources: Bank of America Merrill Lynch; Haver Analytics; Thomson Reuters Datastream; and IMF staff calculations.
Note: Shocks are normalized to a 1 percentage point increase. X-axis units in panel are quarters; t = 0 denotes the quarter of the shock. EMBI = J.P. Morgan Emerging Markets Bond Index.
–0.4 –0.2 0.0 0.2 0.4 0.6 0.8
0 5 10 15 20
–0.2 –0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
0 5 10 15 20
1. Response to 1 percent U.S.
Real GDP Growth Shock
2. Responses from Alternative VAR Specification with Euro Area Real GDP Growth
Baseline specification Alternative specification with euro area real GDP growth
Response to 1 percent U.S. GDP growth shock Response to 1 percent euro area GDP growth shock
–0.06 –0.04 –0.02 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
0 5 10 15 20
–0.2 0.0 0.2 0.4 0.6 0.8
0 5 10 15 20
3. Response to 1 4.
percent China real GDP growth shock (baseline) Response to 1 percent China real investment growth shock (alternative)
Response to 1 percent terms-of-trade growth shock (baseline) Response to 1 percent global commodity price growth shock (alternative) Responses from Baseline and Alternative VAR Specifications
Autoregression Specifications (Percentage points)
Sources: Haver Analytics; IMF, International Financial Statistics database;
Organization for Economic Cooperation and Development; and IMF staff calculations.
Note: Average for all sample economies. Shocks are normalized to a 1 percentage point increase. X-axis units in panels are quarters; t= 0 denotes the quarter of the shock. VAR = vector autoregression.
As Figure 4.24 illustrates, the responses of emerging market economy growth to changes in external condi-tions in the panel VAR are broadly similar to the average responses from the country-speciic SVARs used in the chapter text. he panel VAR typically produces somewhat larger amplitudes, however, such that the cumulated
efects are greater. A 1 percent rise in the U.S. growth rate results in a 0.4 percent rise in emerging market economy growth, whereas a 100 basis point rise in the EMBI yield reduces growth by 0.3 percentage point. However, an increase in China’s growth has a small negative efect on impact, although the efects build up over time.
–1.0 –0.5 0.0 0.5 1.0 1.5
0 5 10 15 20
–0.8 –0.6 –0.4 –0.2
–0.8 –0.6 –0.4
–0.08 –0.04 –0.2
0.0 0.2 0.4 0.6 0.8
0 5 10 15 20
1. Shock to U.S. Real GDP Growth
2. Shock to 10-Year U.S.
Treasury Bond Rate Long sample from 1995:Q1 Baseline sample from 1998:Q1
0.00 0.04 0.08 0.12
0 5 10 15 20
0.0 0.2 0.4
0 5 10 15 20
3. Shock to EMBI Global Yield
4. Shock to Terms-of-Trade Growth
Sources: Haver Analytics; IMF, International Financial Statistics database;
Organization for Economic Cooperation and Development; Thomson Reuters Datastream; and IMF staff calculations.
Note: Shocks are normalized to a 1 percentage point increase. X-axis units in panels are quarters; t = 0 denotes the quarter of the shock.
Figure 4.23. Brazil: Comparison of Responses under the Baseline Model with Responses from Model with Sample Beginning in the First Quarter of 1995
(Percentage points)
–0.5 –0.4 –0.3 –0.2 –0.1 0.0 0.1 0.2
–1.5 –1.2 –0.9 –0.6 –0.3 0.0 0.3 0.6
0 5 10 15 20 –0.4
–0.2 0.0 0.2 0.4 0.6
–1.4 –0.7 0.0 0.7 1.4 2.1
0 5 10 15 20
–0.6 –0.4 –0.2 0.0 0.2 0.4 0.6 0.8
–2.1 –1.4 –0.7 0.0 0.7 1.4 2.1 2.8
0 5 10 15 20
–0.2 –0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6
–0.6 –0.3 0.0 0.3 0.6 0.9 1.2 1.5 1.8
0 5 10 15 20
1. U.S. Real GDP Growth Shock
2. Ten-Year U.S.
Treasury Bond Rate Shock
Baseline specification (VAR, AR(1) prior; left scale) Alternative specification (VAR, white-noise priors; left scale) Alternative specification (panel VAR; right scale)
3. EMBI Global Yield Shock
4. China Real GDP Growth Shock
Figure 4.24. Comparison of Impulse Responses from Panel Vector Autoregression with Responses from the Baseline Model
(Percentage points)
Sources: Haver Analytics; Thomson Reuters Datastream; and IMF staff calculations.
Note: Shocks are normalized to a 1 percentage point increase. X-axis units in panels are quarters; t = 0 denotes the quarter of the shock. AR(1) = first-order autoregression; EMBI = J.P. Morgan Emerging Markets Bond Index; VAR = vector autoregression.
his box uses panel growth regressions to estimate the impact of external demand and global inancial condi-tions on medium-term growth in emerging market economies. hus, it complements the analysis in the chapter, which is more focused on the shorter-term growth implications of changes in external conditions.
Growth regressions, which abstract from the business cycle by aggregating data over ive-year periods, natu-rally lend themselves to addressing questions relating to the medium-term impact of a protracted period of adverse external conditions on emerging market economies’ growth. Also, given wider availability of data at an annual frequency, the indings of the box are applicable to a broader group of emerging markets.
Economic theory suggests several channels through which external conditions afect long-term growth.
he standard growth model is the obvious starting point. Real external shocks, such as an increase in external demand or a change in the terms of trade, directly afect the productivity of capital and therefore capital accumulation.
Financial linkages
As for inancial linkages, arbitrage ensures that a small open economy with an open capital account will be in a steady state when the productivity of domestic capital is equal to the global interest rate. Although there are many reasons why this equalization may never be achieved (for example, country risk, investment costs), an increase in global real interest rates will necessarily reduce funding for marginal investment projects and negatively afect growth. his process can progress in a dramatic fashion, with an increase in international rates precipitating banking crises and the ensuing decrease in output (Eichengreen and Rose, 2004).
his box analyzes the impact of both trade and inancial linkages in a single regression. he two chan-nels operate in opposite directions: whereas a recession in advanced economies may adversely afect emerging market economies’ growth (through a combination of lower external demand and weaker terms of trade), relatively lower interest rates in advanced economy downturns can boost domestic demand growth in emerging markets. Analyzing all external factors simultaneously reduces omitted-variable bias, even if it does not allow identiication of the exogenous impact of each separately.
Specification and methodology
he empirical approach estimates ixed-efects panel growth regressions—for growth averaged over consecu-tive ive-year periods—of the following general form:
∆lnGDPPCi,t = β1'(External Conditions)i,t + β2'Xi,t + γi + ηt + εi,t, (4.1.1) in which
∆lnGDPPCi,t = irst diference in the log of real per capita GDP;
External Conditions = variables measuring external conditions, which include
Trading partner growth, computed following Arora and Vamvakidis (2005),1
Change in the log of the terms of trade, and International inancing conditions (for example, the real interest rate on the 10-year U.S. Treasury bond) interacted with the degree of inancial openness;
Xi,t = standard growth regressors, such as initial level of income, population growth, and investment ratio;
γi = country ixed efect; and
ηt = time ixed efect to control for changes in global conditions not captured by the model.
For most speciications, the panel is estimated for the period 1997–20112 and includes 62 emerging market economies with populations of more than two million, of which 14 are classiied as mineral commod-ity exporters. he emerging market economy universe is larger than the one considered in the chapter, cover-ing a number of countries (mostly in eastern Europe) only recently reclassiied as advanced economies.3
1A similar approach is also used by Drummond and Ramirez (2009) and Dabla-Norris, Espinoza, and Jahan (2012).
2he period is chosen to coincide roughly with the period covered in the chapter. Results, especially those concerning trade linkages, remain broadly unchanged if the period is stretched back to the mid-1980s and even the 1970s.
3he panel is constructed using data from IMF sources (World Economic Outlook, International Financial Statistics, Direction of Trade Statistics, Annual Report on Exchange Arrangements and Exchange Restrictions), as well as from the World Development Indicators (World Bank), Lane and Milesi-Ferretti (2007), Klein and Shambaugh (2008), and the Armed Conlict Dataset (Peace Research Institute Oslo).
he author of this box is Alexander Culiuc.
Trade linkages
he growth regressions are estimated separately for all emerging market economies in the sample and for non–mineral commodity exporters. he regres-sions conirm that emerging markets’ per capita GDP growth is subject to conditional convergence (negative coeicient on lagged GDP per capita), and both investment and the terms of trade have positive growth efects (Table 4.1.1, columns 1 and 2 for the full sample, and columns 3 and 4 for non-commodity-exporting emerging markets). Medium-term growth exhibits a correlation close to one vis-à-vis growth in export partner economies. his elasticity tends to increase with trade openness (column 2 of the table and Figure 4.1.1), particularly for the non-commod-ity-exporting economies (column 4 of the table and Figure 4.1.1). he results also suggest that the terms of trade have a limited role in determining medium-term growth, especially for non–commodity exporters.
he analysis also tracks the relationship between partner growth elasticity and trade openness over time by introducing interaction efects with time dum-mies (Figure 4.1.2). As panel 1 of Figure 4.1.2 shows, partner growth elasticity has been increasing since the
mid-1980s in line with the median export-to-GDP ratio. However, although advanced economy partner growth elasticity has been rising over time, emerg-ing market economy partner growth elasticity started rapidly picking up (from zero) only in the early 1990s (panel 2 of Figure 4.1.2).
he increase in the growth elasticity of emerging markets with respect to growth in their emerging market partners coincides with—and is likely driven by—the growing prominence of Brazil, Russia, India, China, and South Africa (BRICS) and, particularly, the proliferation of supply chains with China. To assess this supposition, the growth regressions are reestimated for all non-BRICS emerging markets (Table 4.1.2 and panels 3 and 4 of Figure 4.1.2).4 Panel 3 of the igure appears to corroborate the hypothesis: for the average emerging market economy, correlation with BRICS growth is fairly high (0.3)
4All partner growth elasticities are weighted by the share of partner countries in the export basket of each emerging market.
his means, among other things, that the BRICS partner growth elasticity is heavily weighted toward China, which, for the aver-age emerging market economy, accounts for more than one-third of exports to the BRICS.
Box 4.1 (continued)
Table 4.1.1. Growth Regressions for Emerging Markets, 1997–2011
All Emerging Market Economies
Non-Commodity-Exporting Emerging Market Economies
(1) (2) (3) (4)
Lagged GDP per Capita (log) –0.053** –0.051** –0.083*** –0.082***
(0.025) (0.025) (0.020) (0.020)
Population Growth 1.473** 1.432** 0.128 0.235
(0.571) (0.542) (0.311) (0.305)
Gross Capital Formation/GDP 0.052 0.062 0.183*** 0.178***
(0.054) (0.058) (0.032) (0.032)
War –0.006 –0.001 0.000 0.000
(0.005) (0.003) (0.003) (0.003)
Terms-of-Trade Growth 0.121* 0.114* 0.066 0.060
(0.068) (0.060) (0.070) (0.068)
Trading Partner GDP Growth 0.910*** 0.692 0.847*** 0.541**
(0.255) (0.466) (0.177) (0.262)
Exports/GDP –0.054 –0.025
(0.043) (0.037)
Trading Partner GDP Growth × Exports/
GDP
0.685 (1.085)
1.072 (1.078)
Time Fixed Effects Yes Yes Yes Yes
Country Fixed Effects Yes Yes Yes Yes
Number of Observations 164 164 121 121
Number of Countries 57 57 42 42
R Squared 0.505 0.486 0.685 0.668
Source: IMF staff calculations.
Note: Standard errors (in parentheses) are clustered at the country level. *, **, *** indicate that coefficients are significant at the 10, 5, and 1 percent levels, respectively.